Regional sociopolitical values are embedded in Large Language Models (LLMs), yet our understanding remains limited regarding how these values are interpreted and expressed by LLMs—particularly as popular models originating from diverse regional, cultural and political contexts increasingly shape public discourse and perceptions of global governance. The Global LLM Values Benchmarking Working Group is undertaking the first systematic, comparative assessment of regional sociopolitical values in LLMs.

Large Language Models like ChatGPT, Claude, DeepSeek and Mistral have become central to global decision-making and communication. However, training these models on regionally biased data often embeds sociopolitical values that influence how global challenges are interpreted and framed. The Global LLMs Values Benchmarking Working Group is conducting a systematic, comparative study to benchmark these values in LLMs from the U.S., China, Europe, India and Russia as well as the Middle East and North Africa. The study examines LLM perspectives on international conflict, climate change, public health and migration. The group will produce tools, papers and convenings to improve AI governance, enhance transparency and foster inclusive AI systems that reflect diverse global perspectives.